#include "xfs_inode.h"
#include "xfs_rw.h"
+kmem_zone_t *xfs_log_ticket_zone;
#define xlog_write_adv_cnt(ptr, len, off, bytes) \
{ (ptr) += (bytes); \
xlog_ticket_t *ticket,
int *continued_write,
int *logoffsetp);
-STATIC void xlog_state_put_ticket(xlog_t *log,
- xlog_ticket_t *tic);
STATIC int xlog_state_release_iclog(xlog_t *log,
xlog_in_core_t *iclog);
STATIC void xlog_state_switch_iclogs(xlog_t *log,
/* local ticket functions */
-STATIC void xlog_state_ticket_alloc(xlog_t *log);
-STATIC xlog_ticket_t *xlog_ticket_get(xlog_t *log,
+STATIC xlog_ticket_t *xlog_ticket_alloc(xlog_t *log,
int unit_bytes,
int count,
char clientid,
uint flags);
-STATIC void xlog_ticket_put(xlog_t *log, xlog_ticket_t *ticket);
#if defined(DEBUG)
STATIC void xlog_verify_dest_ptr(xlog_t *log, __psint_t ptr);
STATIC int xlog_iclogs_empty(xlog_t *log);
#if defined(XFS_LOG_TRACE)
+
+#define XLOG_TRACE_LOGGRANT_SIZE 2048
+#define XLOG_TRACE_ICLOG_SIZE 256
+
+void
+xlog_trace_loggrant_alloc(xlog_t *log)
+{
+ log->l_grant_trace = ktrace_alloc(XLOG_TRACE_LOGGRANT_SIZE, KM_NOFS);
+}
+
+void
+xlog_trace_loggrant_dealloc(xlog_t *log)
+{
+ ktrace_free(log->l_grant_trace);
+}
+
void
xlog_trace_loggrant(xlog_t *log, xlog_ticket_t *tic, xfs_caddr_t string)
{
unsigned long cnts;
- if (!log->l_grant_trace) {
- log->l_grant_trace = ktrace_alloc(2048, KM_NOSLEEP);
- if (!log->l_grant_trace)
- return;
- }
/* ticket counts are 1 byte each */
cnts = ((unsigned long)tic->t_ocnt) | ((unsigned long)tic->t_cnt) << 8;
}
void
+xlog_trace_iclog_alloc(xlog_in_core_t *iclog)
+{
+ iclog->ic_trace = ktrace_alloc(XLOG_TRACE_ICLOG_SIZE, KM_NOFS);
+}
+
+void
+xlog_trace_iclog_dealloc(xlog_in_core_t *iclog)
+{
+ ktrace_free(iclog->ic_trace);
+}
+
+void
xlog_trace_iclog(xlog_in_core_t *iclog, uint state)
{
- if (!iclog->ic_trace)
- iclog->ic_trace = ktrace_alloc(256, KM_SLEEP);
ktrace_enter(iclog->ic_trace,
(void *)((unsigned long)state),
(void *)((unsigned long)current_pid()),
(void *)NULL, (void *)NULL);
}
#else
+
+#define xlog_trace_loggrant_alloc(log)
+#define xlog_trace_loggrant_dealloc(log)
#define xlog_trace_loggrant(log,tic,string)
+
+#define xlog_trace_iclog_alloc(iclog)
+#define xlog_trace_iclog_dealloc(iclog)
#define xlog_trace_iclog(iclog,state)
+
#endif /* XFS_LOG_TRACE */
static void
xlog_grant_add_space_write(struct log *log, int bytes)
{
- log->l_grant_write_bytes += bytes;
- if (log->l_grant_write_bytes > log->l_logsize) {
- log->l_grant_write_bytes -= log->l_logsize;
+ int tmp = log->l_logsize - log->l_grant_write_bytes;
+ if (tmp > bytes)
+ log->l_grant_write_bytes += bytes;
+ else {
log->l_grant_write_cycle++;
+ log->l_grant_write_bytes = bytes - tmp;
}
}
static void
xlog_grant_add_space_reserve(struct log *log, int bytes)
{
- log->l_grant_reserve_bytes += bytes;
- if (log->l_grant_reserve_bytes > log->l_logsize) {
- log->l_grant_reserve_bytes -= log->l_logsize;
+ int tmp = log->l_logsize - log->l_grant_reserve_bytes;
+ if (tmp > bytes)
+ log->l_grant_reserve_bytes += bytes;
+ else {
log->l_grant_reserve_cycle++;
+ log->l_grant_reserve_bytes = bytes - tmp;
}
}
*/
xlog_trace_loggrant(log, ticket, "xfs_log_done: (non-permanent)");
xlog_ungrant_log_space(log, ticket);
- xlog_state_put_ticket(log, ticket);
+ xfs_log_ticket_put(ticket);
} else {
xlog_trace_loggrant(log, ticket, "xfs_log_done: (permanent)");
xlog_regrant_reserve_log_space(log, ticket);
- }
-
- /* If this ticket was a permanent reservation and we aren't
- * trying to release it, reset the inited flags; so next time
- * we write, a start record will be written out.
- */
- if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) &&
- (flags & XFS_LOG_REL_PERM_RESERV) == 0)
+ /* If this ticket was a permanent reservation and we aren't
+ * trying to release it, reset the inited flags; so next time
+ * we write, a start record will be written out.
+ */
ticket->t_flags |= XLOG_TIC_INITED;
+ }
return lsn;
} /* xfs_log_done */
* Asynchronous forces are implemented by setting the WANT_SYNC
* bit in the appropriate in-core log and then returning.
*
- * Synchronous forces are implemented with a semaphore. All callers
- * to force a given lsn to disk will wait on a semaphore attached to the
+ * Synchronous forces are implemented with a signal variable. All callers
+ * to force a given lsn to disk will wait on a the sv attached to the
* specific in-core log. When given in-core log finally completes its
* write to disk, that thread will wake up all threads waiting on the
- * semaphore.
+ * sv.
*/
int
_xfs_log_force(
return xlog_state_sync_all(log, flags, log_flushed);
else
return xlog_state_sync(log, lsn, flags, log_flushed);
-} /* xfs_log_force */
+} /* _xfs_log_force */
+
+/*
+ * Wrapper for _xfs_log_force(), to be used when caller doesn't care
+ * about errors or whether the log was flushed or not. This is the normal
+ * interface to use when trying to unpin items or move the log forward.
+ */
+void
+xfs_log_force(
+ xfs_mount_t *mp,
+ xfs_lsn_t lsn,
+ uint flags)
+{
+ int error;
+ error = _xfs_log_force(mp, lsn, flags, NULL);
+ if (error) {
+ xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: "
+ "error %d returned.", error);
+ }
+}
+
/*
* Attaches a new iclog I/O completion callback routine during
void *iclog_hndl, /* iclog to hang callback off */
xfs_log_callback_t *cb)
{
- xlog_t *log = mp->m_log;
xlog_in_core_t *iclog = (xlog_in_core_t *)iclog_hndl;
int abortflg;
- cb->cb_next = NULL;
- spin_lock(&log->l_icloglock);
+ spin_lock(&iclog->ic_callback_lock);
abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
if (!abortflg) {
ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
*(iclog->ic_callback_tail) = cb;
iclog->ic_callback_tail = &(cb->cb_next);
}
- spin_unlock(&log->l_icloglock);
+ spin_unlock(&iclog->ic_callback_lock);
return abortflg;
} /* xfs_log_notify */
retval = xlog_regrant_write_log_space(log, internal_ticket);
} else {
/* may sleep if need to allocate more tickets */
- internal_ticket = xlog_ticket_get(log, unit_bytes, cnt,
+ internal_ticket = xlog_ticket_alloc(log, unit_bytes, cnt,
client, flags);
+ if (!internal_ticket)
+ return XFS_ERROR(ENOMEM);
internal_ticket->t_trans_type = t_type;
*ticket = internal_ticket;
xlog_trace_loggrant(log, internal_ticket,
}
mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
+ if (!mp->m_log) {
+ cmn_err(CE_WARN, "XFS: Log allocation failed: No memory!");
+ error = ENOMEM;
+ goto out;
+ }
/*
* Initialize the AIL now we have a log.
*/
- spin_lock_init(&mp->m_ail_lock);
error = xfs_trans_ail_init(mp);
if (error) {
cmn_err(CE_WARN, "XFS: AIL initialisation failed: error %d", error);
goto error;
}
+ mp->m_log->l_ailp = mp->m_ail;
/*
* skip log recovery on a norecovery mount. pretend it all
return 0;
error:
xfs_log_unmount_dealloc(mp);
+out:
return error;
} /* xfs_log_mount */
* mp - ubiquitous xfs mount point structure
*/
int
-xfs_log_mount_finish(xfs_mount_t *mp, int mfsi_flags)
+xfs_log_mount_finish(xfs_mount_t *mp)
{
int error;
if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
- error = xlog_recover_finish(mp->m_log, mfsi_flags);
+ error = xlog_recover_finish(mp->m_log);
else {
error = 0;
ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
if (mp->m_flags & XFS_MOUNT_RDONLY)
return 0;
- xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC);
+ error = _xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC, NULL);
+ ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
#ifdef DEBUG
first_iclog = iclog = log->l_iclog;
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
- iclog->ic_refcnt++;
- spin_unlock(&log->l_icloglock);
+ atomic_inc(&iclog->ic_refcnt);
xlog_state_want_sync(log, iclog);
- (void) xlog_state_release_iclog(log, iclog);
+ spin_unlock(&log->l_icloglock);
+ error = xlog_state_release_iclog(log, iclog);
spin_lock(&log->l_icloglock);
if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_DIRTY)) {
if (!XLOG_FORCED_SHUTDOWN(log)) {
- sv_wait(&iclog->ic_forcesema, PMEM,
+ sv_wait(&iclog->ic_force_wait, PMEM,
&log->l_icloglock, s);
} else {
spin_unlock(&log->l_icloglock);
if (tic) {
xlog_trace_loggrant(log, tic, "unmount rec");
xlog_ungrant_log_space(log, tic);
- xlog_state_put_ticket(log, tic);
+ xfs_log_ticket_put(tic);
}
} else {
/*
*/
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
- iclog->ic_refcnt++;
- spin_unlock(&log->l_icloglock);
+ atomic_inc(&iclog->ic_refcnt);
xlog_state_want_sync(log, iclog);
- (void) xlog_state_release_iclog(log, iclog);
+ spin_unlock(&log->l_icloglock);
+ error = xlog_state_release_iclog(log, iclog);
spin_lock(&log->l_icloglock);
|| iclog->ic_state == XLOG_STATE_DIRTY
|| iclog->ic_state == XLOG_STATE_IOERROR) ) {
- sv_wait(&iclog->ic_forcesema, PMEM,
+ sv_wait(&iclog->ic_force_wait, PMEM,
&log->l_icloglock, s);
} else {
spin_unlock(&log->l_icloglock);
}
}
- return 0;
+ return error;
} /* xfs_log_unmount_write */
/*
break;
tail_lsn = 0;
free_bytes -= tic->t_unit_res;
- sv_signal(&tic->t_sema);
+ sv_signal(&tic->t_wait);
tic = tic->t_next;
} while (tic != log->l_write_headq);
}
break;
tail_lsn = 0;
free_bytes -= need_bytes;
- sv_signal(&tic->t_sema);
+ sv_signal(&tic->t_wait);
tic = tic->t_next;
} while (tic != log->l_reserve_headq);
}
int
xfs_log_need_covered(xfs_mount_t *mp)
{
- int needed = 0, gen;
+ int needed = 0;
xlog_t *log = mp->m_log;
if (!xfs_fs_writable(mp))
spin_lock(&log->l_icloglock);
if (((log->l_covered_state == XLOG_STATE_COVER_NEED) ||
(log->l_covered_state == XLOG_STATE_COVER_NEED2))
- && !xfs_trans_first_ail(mp, &gen)
+ && !xfs_trans_ail_tail(log->l_ailp)
&& xlog_iclogs_empty(log)) {
if (log->l_covered_state == XLOG_STATE_COVER_NEED)
log->l_covered_state = XLOG_STATE_COVER_DONE;
xfs_lsn_t tail_lsn;
xlog_t *log = mp->m_log;
- tail_lsn = xfs_trans_tail_ail(mp);
+ tail_lsn = xfs_trans_ail_tail(mp->m_ail);
spin_lock(&log->l_grant_lock);
if (tail_lsn != 0) {
log->l_tail_lsn = tail_lsn;
ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long) 2);
XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
aborted = 0;
-
- /*
- * Some versions of cpp barf on the recursive definition of
- * ic_log -> hic_fields.ic_log and expand ic_log twice when
- * it is passed through two macros. Workaround broken cpp.
- */
l = iclog->ic_log;
/*
- * If the ordered flag has been removed by a lower
- * layer, it means the underlyin device no longer supports
+ * If the _XFS_BARRIER_FAILED flag was set by a lower
+ * layer, it means the underlying device no longer supports
* barrier I/O. Warn loudly and turn off barriers.
*/
- if ((l->l_mp->m_flags & XFS_MOUNT_BARRIER) && !XFS_BUF_ORDERED(bp)) {
+ if (bp->b_flags & _XFS_BARRIER_FAILED) {
+ bp->b_flags &= ~_XFS_BARRIER_FAILED;
l->l_mp->m_flags &= ~XFS_MOUNT_BARRIER;
xfs_fs_cmn_err(CE_WARN, l->l_mp,
"xlog_iodone: Barriers are no longer supported"
int i;
int iclogsize;
- log = (xlog_t *)kmem_zalloc(sizeof(xlog_t), KM_SLEEP);
+ log = kmem_zalloc(sizeof(xlog_t), KM_MAYFAIL);
+ if (!log)
+ return NULL;
log->l_mp = mp;
log->l_targ = log_target;
xlog_get_iclog_buffer_size(mp, log);
bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp);
+ if (!bp)
+ goto out_free_log;
XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
spin_lock_init(&log->l_icloglock);
spin_lock_init(&log->l_grant_lock);
- initnsema(&log->l_flushsema, 0, "ic-flush");
- xlog_state_ticket_alloc(log); /* wait until after icloglock inited */
+ sv_init(&log->l_flush_wait, 0, "flush_wait");
+ xlog_trace_loggrant_alloc(log);
/* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0);
iclogsize = log->l_iclog_size;
ASSERT(log->l_iclog_size >= 4096);
for (i=0; i < log->l_iclog_bufs; i++) {
- *iclogp = (xlog_in_core_t *)
- kmem_zalloc(sizeof(xlog_in_core_t), KM_SLEEP);
+ *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
+ if (!*iclogp)
+ goto out_free_iclog;
+
iclog = *iclogp;
iclog->ic_prev = prev_iclog;
prev_iclog = iclog;
bp = xfs_buf_get_noaddr(log->l_iclog_size, mp->m_logdev_targp);
+ if (!bp)
+ goto out_free_iclog;
if (!XFS_BUF_CPSEMA(bp))
ASSERT(0);
XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
iclog->ic_bp = bp;
- iclog->hic_data = bp->b_addr;
-
+ iclog->ic_data = bp->b_addr;
+#ifdef DEBUG
log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
-
+#endif
head = &iclog->ic_header;
memset(head, 0, sizeof(xlog_rec_header_t));
head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
head->h_fmt = cpu_to_be32(XLOG_FMT);
memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
-
iclog->ic_size = XFS_BUF_SIZE(bp) - log->l_iclog_hsize;
iclog->ic_state = XLOG_STATE_ACTIVE;
iclog->ic_log = log;
+ atomic_set(&iclog->ic_refcnt, 0);
+ spin_lock_init(&iclog->ic_callback_lock);
iclog->ic_callback_tail = &(iclog->ic_callback);
- iclog->ic_datap = (char *)iclog->hic_data + log->l_iclog_hsize;
+ iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp));
ASSERT(XFS_BUF_VALUSEMA(iclog->ic_bp) <= 0);
- sv_init(&iclog->ic_forcesema, SV_DEFAULT, "iclog-force");
- sv_init(&iclog->ic_writesema, SV_DEFAULT, "iclog-write");
+ sv_init(&iclog->ic_force_wait, SV_DEFAULT, "iclog-force");
+ sv_init(&iclog->ic_write_wait, SV_DEFAULT, "iclog-write");
+
+ xlog_trace_iclog_alloc(iclog);
iclogp = &iclog->ic_next;
}
log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
return log;
+
+out_free_iclog:
+ for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
+ prev_iclog = iclog->ic_next;
+ if (iclog->ic_bp) {
+ sv_destroy(&iclog->ic_force_wait);
+ sv_destroy(&iclog->ic_write_wait);
+ xfs_buf_free(iclog->ic_bp);
+ xlog_trace_iclog_dealloc(iclog);
+ }
+ kmem_free(iclog);
+ }
+ spinlock_destroy(&log->l_icloglock);
+ spinlock_destroy(&log->l_grant_lock);
+ xlog_trace_loggrant_dealloc(log);
+ xfs_buf_free(log->l_xbuf);
+out_free_log:
+ kmem_free(log);
+ return NULL;
} /* xlog_alloc_log */
*/
if (threshold_lsn &&
!XLOG_FORCED_SHUTDOWN(log))
- xfs_trans_push_ail(mp, threshold_lsn);
+ xfs_trans_ail_push(log->l_ailp, threshold_lsn);
} /* xlog_grant_push_ail */
int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
XFS_STATS_INC(xs_log_writes);
- ASSERT(iclog->ic_refcnt == 0);
+ ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
/* Add for LR header */
count_init = log->l_iclog_hsize + iclog->ic_offset;
xlog_dealloc_log(xlog_t *log)
{
xlog_in_core_t *iclog, *next_iclog;
- xlog_ticket_t *tic, *next_tic;
int i;
iclog = log->l_iclog;
for (i=0; i<log->l_iclog_bufs; i++) {
- sv_destroy(&iclog->ic_forcesema);
- sv_destroy(&iclog->ic_writesema);
+ sv_destroy(&iclog->ic_force_wait);
+ sv_destroy(&iclog->ic_write_wait);
xfs_buf_free(iclog->ic_bp);
-#ifdef XFS_LOG_TRACE
- if (iclog->ic_trace != NULL) {
- ktrace_free(iclog->ic_trace);
- }
-#endif
+ xlog_trace_iclog_dealloc(iclog);
next_iclog = iclog->ic_next;
- kmem_free(iclog, sizeof(xlog_in_core_t));
+ kmem_free(iclog);
iclog = next_iclog;
}
- freesema(&log->l_flushsema);
spinlock_destroy(&log->l_icloglock);
spinlock_destroy(&log->l_grant_lock);
- /* XXXsup take a look at this again. */
- if ((log->l_ticket_cnt != log->l_ticket_tcnt) &&
- !XLOG_FORCED_SHUTDOWN(log)) {
- xfs_fs_cmn_err(CE_WARN, log->l_mp,
- "xlog_dealloc_log: (cnt: %d, total: %d)",
- log->l_ticket_cnt, log->l_ticket_tcnt);
- /* ASSERT(log->l_ticket_cnt == log->l_ticket_tcnt); */
-
- } else {
- tic = log->l_unmount_free;
- while (tic) {
- next_tic = tic->t_next;
- kmem_free(tic, PAGE_SIZE);
- tic = next_tic;
- }
- }
xfs_buf_free(log->l_xbuf);
-#ifdef XFS_LOG_TRACE
- if (log->l_trace != NULL) {
- ktrace_free(log->l_trace);
- }
- if (log->l_grant_trace != NULL) {
- ktrace_free(log->l_grant_trace);
- }
-#endif
+ xlog_trace_loggrant_dealloc(log);
log->l_mp->m_log = NULL;
- kmem_free(log, sizeof(xlog_t));
+ kmem_free(log);
} /* xlog_dealloc_log */
/*
if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
record_cnt = data_cnt = 0;
+ spin_lock(&log->l_icloglock);
xlog_state_want_sync(log, iclog);
+ spin_unlock(&log->l_icloglock);
if (commit_iclog) {
ASSERT(flags & XLOG_COMMIT_TRANS);
*commit_iclog = iclog;
/* Clean iclogs starting from the head. This ordering must be
* maintained, so an iclog doesn't become ACTIVE beyond one that
* is SYNCING. This is also required to maintain the notion that we use
- * a counting semaphore to hold off would be writers to the log when every
+ * a ordered wait queue to hold off would be writers to the log when every
* iclog is trying to sync to disk.
*
* State Change: DIRTY -> ACTIVE
if (iclog->ic_state == XLOG_STATE_DIRTY) {
iclog->ic_state = XLOG_STATE_ACTIVE;
iclog->ic_offset = 0;
- iclog->ic_callback = NULL; /* don't need to free */
+ ASSERT(iclog->ic_callback == NULL);
/*
* If the number of ops in this iclog indicate it just
* contains the dummy transaction, we can
int funcdidcallbacks; /* flag: function did callbacks */
int repeats; /* for issuing console warnings if
* looping too many times */
+ int wake = 0;
spin_lock(&log->l_icloglock);
first_iclog = iclog = log->l_iclog;
be64_to_cpu(iclog->ic_header.h_lsn);
spin_unlock(&log->l_grant_lock);
- /*
- * Keep processing entries in the callback list
- * until we come around and it is empty. We
- * need to atomically see that the list is
- * empty and change the state to DIRTY so that
- * we don't miss any more callbacks being added.
- */
- spin_lock(&log->l_icloglock);
} else {
+ spin_unlock(&log->l_icloglock);
ioerrors++;
}
- cb = iclog->ic_callback;
+ /*
+ * Keep processing entries in the callback list until
+ * we come around and it is empty. We need to
+ * atomically see that the list is empty and change the
+ * state to DIRTY so that we don't miss any more
+ * callbacks being added.
+ */
+ spin_lock(&iclog->ic_callback_lock);
+ cb = iclog->ic_callback;
while (cb) {
iclog->ic_callback_tail = &(iclog->ic_callback);
iclog->ic_callback = NULL;
- spin_unlock(&log->l_icloglock);
+ spin_unlock(&iclog->ic_callback_lock);
/* perform callbacks in the order given */
for (; cb; cb = cb_next) {
cb_next = cb->cb_next;
cb->cb_func(cb->cb_arg, aborted);
}
- spin_lock(&log->l_icloglock);
+ spin_lock(&iclog->ic_callback_lock);
cb = iclog->ic_callback;
}
loopdidcallbacks++;
funcdidcallbacks++;
+ spin_lock(&log->l_icloglock);
ASSERT(iclog->ic_callback == NULL);
+ spin_unlock(&iclog->ic_callback_lock);
if (!(iclog->ic_state & XLOG_STATE_IOERROR))
iclog->ic_state = XLOG_STATE_DIRTY;
xlog_state_clean_log(log);
/* wake up threads waiting in xfs_log_force() */
- sv_broadcast(&iclog->ic_forcesema);
+ sv_broadcast(&iclog->ic_force_wait);
iclog = iclog->ic_next;
} while (first_iclog != iclog);
repeats = 0;
xfs_fs_cmn_err(CE_WARN, log->l_mp,
"%s: possible infinite loop (%d iterations)",
- __FUNCTION__, flushcnt);
+ __func__, flushcnt);
}
} while (!ioerrors && loopdidcallbacks);
}
#endif
- flushcnt = 0;
- if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) {
- flushcnt = log->l_flushcnt;
- log->l_flushcnt = 0;
- }
+ if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
+ wake = 1;
spin_unlock(&log->l_icloglock);
- while (flushcnt--)
- vsema(&log->l_flushsema);
-} /* xlog_state_do_callback */
+
+ if (wake)
+ sv_broadcast(&log->l_flush_wait);
+}
/*
* the second completion goes through.
*
* Callbacks could take time, so they are done outside the scope of the
- * global state machine log lock. Assume that the calls to cvsema won't
- * take a long time. At least we know it won't sleep.
+ * global state machine log lock.
*/
STATIC void
xlog_state_done_syncing(
ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
iclog->ic_state == XLOG_STATE_IOERROR);
- ASSERT(iclog->ic_refcnt == 0);
+ ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
* iclog buffer, we wake them all, one will get to do the
* I/O, the others get to wait for the result.
*/
- sv_broadcast(&iclog->ic_writesema);
+ sv_broadcast(&iclog->ic_write_wait);
spin_unlock(&log->l_icloglock);
xlog_state_do_callback(log, aborted, iclog); /* also cleans log */
} /* xlog_state_done_syncing */
/*
* If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
- * sleep. The flush semaphore is set to the number of in-core buffers and
- * decremented around disk syncing. Therefore, if all buffers are syncing,
- * this semaphore will cause new writes to sleep until a sync completes.
- * Otherwise, this code just does p() followed by v(). This approximates
- * a sleep/wakeup except we can't race.
+ * sleep. We wait on the flush queue on the head iclog as that should be
+ * the first iclog to complete flushing. Hence if all iclogs are syncing,
+ * we will wait here and all new writes will sleep until a sync completes.
*
* The in-core logs are used in a circular fashion. They are not used
* out-of-order even when an iclog past the head is free.
}
iclog = log->l_iclog;
- if (! (iclog->ic_state == XLOG_STATE_ACTIVE)) {
- log->l_flushcnt++;
- spin_unlock(&log->l_icloglock);
+ if (iclog->ic_state != XLOG_STATE_ACTIVE) {
xlog_trace_iclog(iclog, XLOG_TRACE_SLEEP_FLUSH);
XFS_STATS_INC(xs_log_noiclogs);
- /* Ensure that log writes happen */
- psema(&log->l_flushsema, PINOD);
+
+ /* Wait for log writes to have flushed */
+ sv_wait(&log->l_flush_wait, 0, &log->l_icloglock, 0);
goto restart;
}
- ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
+
head = &iclog->ic_header;
- iclog->ic_refcnt++; /* prevents sync */
+ atomic_inc(&iclog->ic_refcnt); /* prevents sync */
log_offset = iclog->ic_offset;
/* On the 1st write to an iclog, figure out lsn. This works
if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
- /* If I'm the only one writing to this iclog, sync it to disk */
- if (iclog->ic_refcnt == 1) {
+ /*
+ * If I'm the only one writing to this iclog, sync it to disk.
+ * We need to do an atomic compare and decrement here to avoid
+ * racing with concurrent atomic_dec_and_lock() calls in
+ * xlog_state_release_iclog() when there is more than one
+ * reference to the iclog.
+ */
+ if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
+ /* we are the only one */
spin_unlock(&log->l_icloglock);
- if ((error = xlog_state_release_iclog(log, iclog)))
+ error = xlog_state_release_iclog(log, iclog);
+ if (error)
return error;
} else {
- iclog->ic_refcnt--;
spin_unlock(&log->l_icloglock);
}
goto restart;
goto error_return;
XFS_STATS_INC(xs_sleep_logspace);
- sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
+ sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
/*
* If we got an error, and the filesystem is shutting down,
* we'll catch it down below. So just continue...
xlog_trace_loggrant(log, tic,
"xlog_grant_log_space: sleep 2");
XFS_STATS_INC(xs_sleep_logspace);
- sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
+ sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
if (XLOG_FORCED_SHUTDOWN(log)) {
spin_lock(&log->l_grant_lock);
if (free_bytes < ntic->t_unit_res)
break;
free_bytes -= ntic->t_unit_res;
- sv_signal(&ntic->t_sema);
+ sv_signal(&ntic->t_wait);
ntic = ntic->t_next;
} while (ntic != log->l_write_headq);
xlog_trace_loggrant(log, tic,
"xlog_regrant_write_log_space: sleep 1");
XFS_STATS_INC(xs_sleep_logspace);
- sv_wait(&tic->t_sema, PINOD|PLTWAIT,
+ sv_wait(&tic->t_wait, PINOD|PLTWAIT,
&log->l_grant_lock, s);
/* If we're shutting down, this tic is already
if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
xlog_ins_ticketq(&log->l_write_headq, tic);
XFS_STATS_INC(xs_sleep_logspace);
- sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
+ sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
/* If we're shutting down, this tic is already off the queue */
if (XLOG_FORCED_SHUTDOWN(log)) {
/*
- * Atomically put back used ticket.
- */
-STATIC void
-xlog_state_put_ticket(xlog_t *log,
- xlog_ticket_t *tic)
-{
- spin_lock(&log->l_icloglock);
- xlog_ticket_put(log, tic);
- spin_unlock(&log->l_icloglock);
-} /* xlog_state_put_ticket */
-
-/*
* Flush iclog to disk if this is the last reference to the given iclog and
* the WANT_SYNC bit is set.
*
{
int sync = 0; /* do we sync? */
- spin_lock(&log->l_icloglock);
+ if (iclog->ic_state & XLOG_STATE_IOERROR)
+ return XFS_ERROR(EIO);
+
+ ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
+ if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
+ return 0;
+
if (iclog->ic_state & XLOG_STATE_IOERROR) {
spin_unlock(&log->l_icloglock);
return XFS_ERROR(EIO);
}
-
- ASSERT(iclog->ic_refcnt > 0);
ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_WANT_SYNC);
- if (--iclog->ic_refcnt == 0 &&
- iclog->ic_state == XLOG_STATE_WANT_SYNC) {
+ if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
/* update tail before writing to iclog */
xlog_assign_tail_lsn(log->l_mp);
sync++;
* 2. the current iclog is drity, and the previous iclog is in the
* active or dirty state.
*
- * We may sleep (call psema) if:
+ * We may sleep if:
*
* 1. the current iclog is not in the active nor dirty state.
* 2. the current iclog dirty, and the previous iclog is not in the
* previous iclog and go to sleep.
*/
if (iclog->ic_state == XLOG_STATE_DIRTY ||
- (iclog->ic_refcnt == 0 && iclog->ic_offset == 0)) {
+ (atomic_read(&iclog->ic_refcnt) == 0
+ && iclog->ic_offset == 0)) {
iclog = iclog->ic_prev;
if (iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_DIRTY)
else
goto maybe_sleep;
} else {
- if (iclog->ic_refcnt == 0) {
+ if (atomic_read(&iclog->ic_refcnt) == 0) {
/* We are the only one with access to this
* iclog. Flush it out now. There should
* be a roundoff of zero to show that someone
* has already taken care of the roundoff from
* the previous sync.
*/
- iclog->ic_refcnt++;
+ atomic_inc(&iclog->ic_refcnt);
lsn = be64_to_cpu(iclog->ic_header.h_lsn);
xlog_state_switch_iclogs(log, iclog, 0);
spin_unlock(&log->l_icloglock);
return XFS_ERROR(EIO);
}
XFS_STATS_INC(xs_log_force_sleep);
- sv_wait(&iclog->ic_forcesema, PINOD, &log->l_icloglock, s);
+ sv_wait(&iclog->ic_force_wait, PINOD, &log->l_icloglock, s);
/*
* No need to grab the log lock here since we're
* only deciding whether or not to return EIO
XLOG_STATE_SYNCING))) {
ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
XFS_STATS_INC(xs_log_force_sleep);
- sv_wait(&iclog->ic_prev->ic_writesema, PSWP,
+ sv_wait(&iclog->ic_prev->ic_write_wait, PSWP,
&log->l_icloglock, s);
*log_flushed = 1;
already_slept = 1;
goto try_again;
} else {
- iclog->ic_refcnt++;
+ atomic_inc(&iclog->ic_refcnt);
xlog_state_switch_iclogs(log, iclog, 0);
spin_unlock(&log->l_icloglock);
if (xlog_state_release_iclog(log, iclog))
!(iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
/*
- * Don't wait on the forcesema if we know that we've
+ * Don't wait on completion if we know that we've
* gotten a log write error.
*/
if (iclog->ic_state & XLOG_STATE_IOERROR) {
return XFS_ERROR(EIO);
}
XFS_STATS_INC(xs_log_force_sleep);
- sv_wait(&iclog->ic_forcesema, PSWP, &log->l_icloglock, s);
+ sv_wait(&iclog->ic_force_wait, PSWP, &log->l_icloglock, s);
/*
* No need to grab the log lock here since we're
* only deciding whether or not to return EIO
STATIC void
xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog)
{
- spin_lock(&log->l_icloglock);
+ ASSERT(spin_is_locked(&log->l_icloglock));
if (iclog->ic_state == XLOG_STATE_ACTIVE) {
xlog_state_switch_iclogs(log, iclog, 0);
ASSERT(iclog->ic_state &
(XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
}
-
- spin_unlock(&log->l_icloglock);
-} /* xlog_state_want_sync */
-
+}
/*****************************************************************************
*/
/*
- * Algorithm doesn't take into account page size. ;-(
+ * Free a used ticket when it's refcount falls to zero.
*/
-STATIC void
-xlog_state_ticket_alloc(xlog_t *log)
+void
+xfs_log_ticket_put(
+ xlog_ticket_t *ticket)
{
- xlog_ticket_t *t_list;
- xlog_ticket_t *next;
- xfs_caddr_t buf;
- uint i = (PAGE_SIZE / sizeof(xlog_ticket_t)) - 2;
-
- /*
- * The kmem_zalloc may sleep, so we shouldn't be holding the
- * global lock. XXXmiken: may want to use zone allocator.
- */
- buf = (xfs_caddr_t) kmem_zalloc(PAGE_SIZE, KM_SLEEP);
-
- spin_lock(&log->l_icloglock);
-
- /* Attach 1st ticket to Q, so we can keep track of allocated memory */
- t_list = (xlog_ticket_t *)buf;
- t_list->t_next = log->l_unmount_free;
- log->l_unmount_free = t_list++;
- log->l_ticket_cnt++;
- log->l_ticket_tcnt++;
-
- /* Next ticket becomes first ticket attached to ticket free list */
- if (log->l_freelist != NULL) {
- ASSERT(log->l_tail != NULL);
- log->l_tail->t_next = t_list;
- } else {
- log->l_freelist = t_list;
+ ASSERT(atomic_read(&ticket->t_ref) > 0);
+ if (atomic_dec_and_test(&ticket->t_ref)) {
+ sv_destroy(&ticket->t_wait);
+ kmem_zone_free(xfs_log_ticket_zone, ticket);
}
- log->l_ticket_cnt++;
- log->l_ticket_tcnt++;
-
- /* Cycle through rest of alloc'ed memory, building up free Q */
- for ( ; i > 0; i--) {
- next = t_list + 1;
- t_list->t_next = next;
- t_list = next;
- log->l_ticket_cnt++;
- log->l_ticket_tcnt++;
- }
- t_list->t_next = NULL;
- log->l_tail = t_list;
- spin_unlock(&log->l_icloglock);
-} /* xlog_state_ticket_alloc */
-
+}
-/*
- * Put ticket into free list
- *
- * Assumption: log lock is held around this call.
- */
-STATIC void
-xlog_ticket_put(xlog_t *log,
- xlog_ticket_t *ticket)
+xlog_ticket_t *
+xfs_log_ticket_get(
+ xlog_ticket_t *ticket)
{
- sv_destroy(&ticket->t_sema);
-
- /*
- * Don't think caching will make that much difference. It's
- * more important to make debug easier.
- */
-#if 0
- /* real code will want to use LIFO for caching */
- ticket->t_next = log->l_freelist;
- log->l_freelist = ticket;
- /* no need to clear fields */
-#else
- /* When we debug, it is easier if tickets are cycled */
- ticket->t_next = NULL;
- if (log->l_tail) {
- log->l_tail->t_next = ticket;
- } else {
- ASSERT(log->l_freelist == NULL);
- log->l_freelist = ticket;
- }
- log->l_tail = ticket;
-#endif /* DEBUG */
- log->l_ticket_cnt++;
-} /* xlog_ticket_put */
-
+ ASSERT(atomic_read(&ticket->t_ref) > 0);
+ atomic_inc(&ticket->t_ref);
+ return ticket;
+}
/*
- * Grab ticket off freelist or allocation some more
+ * Allocate and initialise a new log ticket.
*/
STATIC xlog_ticket_t *
-xlog_ticket_get(xlog_t *log,
+xlog_ticket_alloc(xlog_t *log,
int unit_bytes,
int cnt,
char client,
xlog_ticket_t *tic;
uint num_headers;
- alloc:
- if (log->l_freelist == NULL)
- xlog_state_ticket_alloc(log); /* potentially sleep */
-
- spin_lock(&log->l_icloglock);
- if (log->l_freelist == NULL) {
- spin_unlock(&log->l_icloglock);
- goto alloc;
- }
- tic = log->l_freelist;
- log->l_freelist = tic->t_next;
- if (log->l_freelist == NULL)
- log->l_tail = NULL;
- log->l_ticket_cnt--;
- spin_unlock(&log->l_icloglock);
+ tic = kmem_zone_zalloc(xfs_log_ticket_zone, KM_SLEEP|KM_MAYFAIL);
+ if (!tic)
+ return NULL;
/*
* Permanent reservations have up to 'cnt'-1 active log operations
unit_bytes += 2*BBSIZE;
}
+ atomic_set(&tic->t_ref, 1);
tic->t_unit_res = unit_bytes;
tic->t_curr_res = unit_bytes;
tic->t_cnt = cnt;
tic->t_trans_type = 0;
if (xflags & XFS_LOG_PERM_RESERV)
tic->t_flags |= XLOG_TIC_PERM_RESERV;
- sv_init(&(tic->t_sema), SV_DEFAULT, "logtick");
+ sv_init(&(tic->t_wait), SV_DEFAULT, "logtick");
xlog_tic_reset_res(tic);
return tic;
-} /* xlog_ticket_get */
+}
/******************************************************************************
ptr = iclog->ic_datap;
base_ptr = ptr;
ophead = (xlog_op_header_t *)ptr;
- xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
+ xhdr = iclog->ic_data;
for (i = 0; i < len; i++) {
ophead = (xlog_op_header_t *)ptr;
if (!log ||
log->l_flags & XLOG_ACTIVE_RECOVERY) {
mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
- XFS_BUF_DONE(mp->m_sb_bp);
+ if (mp->m_sb_bp)
+ XFS_BUF_DONE(mp->m_sb_bp);
return 0;
}
* before we mark the filesystem SHUTDOWN and wake
* everybody up to tell the bad news.
*/
- spin_lock(&log->l_grant_lock);
spin_lock(&log->l_icloglock);
+ spin_lock(&log->l_grant_lock);
mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
- XFS_BUF_DONE(mp->m_sb_bp);
+ if (mp->m_sb_bp)
+ XFS_BUF_DONE(mp->m_sb_bp);
+
/*
* This flag is sort of redundant because of the mount flag, but
* it's good to maintain the separation between the log and the rest
*/
if ((tic = log->l_reserve_headq)) {
do {
- sv_signal(&tic->t_sema);
+ sv_signal(&tic->t_wait);
tic = tic->t_next;
} while (tic != log->l_reserve_headq);
}
if ((tic = log->l_write_headq)) {
do {
- sv_signal(&tic->t_sema);
+ sv_signal(&tic->t_wait);
tic = tic->t_next;
} while (tic != log->l_write_headq);
}
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff